PNG images not being deinterlaced by read_image

[ConradStack]

🐛 Bug

Interlaced PNG files are not de-interlaced when loaded by read_image.

To Reproduce

Steps to reproduce the behavior:

1. Download test image (attached):

# download image
wget https://imagemagick.org/image/wizard.png

# create interlaced version using imagemagick 
convert -strip -interlace PNG wizard.png wizard-interlaced.png

# print imagemagick version info
convert -version

Version: ImageMagick 7.1.0-2 Q16 x86_64 2021-06-25 https://imagemagick.org
Copyright: (C) 1999-2021 ImageMagick Studio LLC
License: https://imagemagick.org/script/license.php
Features: Cipher DPC HDRI Modules OpenMP(5.0)
Delegates (built-in): bzlib fontconfig freetype gslib heic jng jp2 jpeg lcms lqr ltdl lzma openexr png ps tiff webp xml zlib

2. Compare read_image to PIL.Image.open

import numpy as np
from PIL import Image
from torchvision.io import read_image, ImageReadMode
from torchvision.transforms import ToPILImage

# Using PIL:
pil_notinterlaced = Image.open("wizard.png")
pil_interlaced = Image.open("wizard-interlaced.png")
# PASS
assert np.all( np.asarray(pil_notinterlaced) == np.asarray(pil_interlaced) ), "differences btw interlaced and non-interlaced images loaded with PIL"


# Using torchvision:
tv_notinterlaced = read_image("wizard.png")
tv_interlaced = read_image("wizard-interlaced.png")

# PASS
assert tv_notinterlaced.shape == tv_interlaced.shape, "not the same shape"

# FAIL
assert (tv_notinterlaced == tv_interlaced).all(), "differences btw interlaced and non-interlaced images loaded with torchvision"

# ... (for manual comparison) convert the interlaced image tensor to PIL and save
topil = ToPILImage()
topil(tv_interlaced).save("wizard-interlaced-savedfromtv.png")

wizard.png:

wizard-interlaced.png:

wizard-interlaced-savedfromtv.png:

Expected behavior

Using read_image to load an interlaced PNG image would automatically de-interlace it.

Environment

Collecting environment information...
PyTorch version: 1.9.0
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A

OS: macOS 11.4 (x86_64)
GCC version: Could not collect
Clang version: 12.0.0 (clang-1200.0.32.2)
CMake version: version 3.17.3
Libc version: N/A

Python version: 3.8.8 (default, Feb 24 2021, 13:46:16)  [Clang 10.0.0 ] (64-bit runtime)
Python platform: macOS-10.16-x86_64-i386-64bit
Is CUDA available: False
CUDA runtime version: No CUDA
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
HIP runtime version: N/A
MIOpen runtime version: N/A

Versions of relevant libraries:
[pip3] numpy==1.21.0
[pip3] torch==1.9.0
[pip3] torchtext==0.10.0
[pip3] torchvision==0.10.0
[conda] blas                      1.0                         mkl  
[conda] ffmpeg                    4.3                  h0a44026_0    pytorch
[conda] mkl                       2021.3.0           hecd8cb5_517  
[conda] mkl-service               2.4.0            py38h9ed2024_0  
[conda] mkl_fft                   1.3.0            py38h4a7008c_2  
[conda] mkl_random                1.2.2            py38hb2f4e1b_0  
[conda] numpy                     1.21.0                   pypi_0    pypi
[conda] numpy-base                1.20.3           py38he0bd621_0  
[conda] pytorch                   1.9.0                   py3.8_0    pytorch
[conda] torchtext                 0.10.0                     py38    pytorch
[conda] torchvision               0.10.0                 py38_cpu    pytorch

Additional context

I reproduced the issue handling interlaced PNGs on a linux machine running centos 7 w/ similar python specs and using other interlaced PNG images that had not been converted by imagemagick.

[fmassa]

Thanks for the report!

We will look into it and fix this problem